AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Both Style and Fog Matter: Cumulative Domain Adaptation for Semantic Foggy Scene Understanding

Xianzheng Ma, Zhixiang Wang, Yacheng Zhan, Yinqiang Zheng, Zheng Wang, Dengxin Dai, Chia-Wen Lin

CVPR 2022 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Domain-generalized segmentation · 52 citations (OpenAlex)

Check your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Decompose transfer into style then fog adaptation rather than adapting directly (ablation)CuDA-Net · Foggy Zurich-testmIoU: 42.49 vs 40.21Improved
Add style adaptation (ablation)DeepLab-v2 fog-domain segmentation · Foggy Zurich-testmIoU: 25.89 -> 39.16Improved
Add fog adaptation after style adaptation (ablation)CuDA-Net · Foggy Zurich-testmIoU: 39.16 -> 42.49; +3.33 mIoUImproved
Add direct dual-factor adaptation after style and fog adaptation (ablation)CuDA-Net · Foggy Zurich-testmIoU: 42.49 -> 43.06Improved
Use cyclical training with T=2 rather than T=1 (ablation)CuDA-Net without cumulative loss · Foggy Zurich-testmIoU: 45.32 -> 45.78Improved
Vary cyclical training from T=1 to T=3 (ablation)CuDA-Net without cumulative loss · Foggy Zurich-testmIoU: 45.32; 45.78; 45.45No clear effect
Use L2 rather than L1 or cosine distance in cumulative loss (ablation)CuDA-Net · Foggy Zurich-testmIoU: 47.64; 47.23; 48.21Improved
Add dense synthetic foggy images to trainingCuDA-Net · Foggy ZurichmIoU: 48.2 -> 49.1; 0.9%Improved
Use two-step adaptation instead of direct adaptation (ablation)CuDA-Net · ACDC rain validationmIoU: 46.2 -> 48.5Improved
Use two-step adaptation instead of direct adaptation (ablation)CuDA-Net · ACDC snow validationmIoU: 44.8 -> 47.2Improved
Select intermediate-domain images manually rather than with a CNN (ablation)CuDA-Net · Foggy Zurich-testmIoU: CNN-based: 47.5, 47.5, 47.2; manual: 48.0, 48.2, 48.2Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy